darknet - forward-yolo-layer函数

static int entry_index(layer l, int batch, int location, int entry)
{ 
    int n =   location / (l.w*l.h);
    int loc = location % (l.w*l.h);
    return batch*l.outputs + n*l.w*l.h*(4+l.classes+1) + entry*l.w*l.h + loc;
}

box get_yolo_box(float *x, float *biases, int n, int index, int i, int j, int lw, int lh, int w, int h, int stride)
{
//(w,h) 输入图片尺寸 (lw,lh)当前特征图尺寸
    box b;
    b.x = (i + x[index + 0*stride]) / lw;
    b.y = (j + x[index + 1*stride]) / lh;
    b.w = exp(x[index + 2*stride]) * biases[2*n]   / w;
    b.h = exp(x[index + 3*stride]) * biases[2*n+1] / h;
    return b;
}

void forward_yolo_layer(const layer l, network_state state)
{
    int i,j,b,t,n;
    memcpy(l.output, state.input, l.outputs*l.batch*sizeof(float));

#ifndef GPU
    for (b = 0; b < l.batch; ++b){
        for(n = 0; n < l.n; ++n){
            int index = entry_index(l, b, n*l.w*l.h, 0);
            activate_array(l.output + index, 2*l.w*l.h, LOGISTIC);
            index = entry_index(l, b, n*l.w*l.h, 4);
            activate_array(l.output + index, (1+l.classes)*l.w*l.h, LOGISTIC);
        }
    }
#endif

    memset(l.delta, 0, l.outputs * l.batch * sizeof(float));
    if(!state.train) return; //不做训练时,直接退出,以下都是训练代码
    float avg_iou = 0;
    float recall = 0;
    float recall75 = 0;
    float avg_cat = 0;
    float avg_obj = 0;
    float avg_anyobj = 0;
    int count = 0;
    int class_count = 0;
    *(l.cost) = 0;
    for (b = 0; b < l.batch; ++b) { //遍历batch
        for (j = 0; j < l.h; ++j) { //遍历h
            for (i = 0; i < l.w; ++i) { //遍历w
                for (n = 0; n < l.n; ++n) { //遍历指定的anchor(不超过l.total)
                    //内存布局: batch-anchor-xoffset-yoffset-w-h-objectness-classid
                    //xoffset,yoffset,bw,bh,objectness,classid的尺寸都是l.w * l.h
                    int box_index = entry_index(l, b, n*l.w*l.h + j*l.w + i, 0);
                    //(i,j) 对应的第l.mask[n]个anchor的预测结果
                    box pred = get_yolo_box(l.output, l.biases, l.mask[n], box_index, i, j, l.w, l.h, state.net.w, state.net.h, l.w*l.h);
                    float best_iou = 0;
                    int best_t = 0;
                    //遍历图中所有groundtruth object ,找出和pred重合度最高的一个object
                    for(t = 0; t < l.max_boxes; ++t){ //max_boxes: cfg中配置,默认90,单个图片中目标个数最大值
                        box truth = float_to_box_stride(state.truth + t*(4 + 1) + b*l.truths, 1); //获得第t个groundtruth objec的类别
                        int class_id = state.truth[t*(4 + 1) + b*l.truths + 4]; //第t个groundtruth objec的类别
                        if (class_id >= l.classes) {
                            printf(" Warning: in txt-labels class_id=%d >= classes=%d in cfg-file. In txt-labels class_id should be [from 0 to %d] \n", class_id, l.classes, l.classes - 1);
                            getchar();
                            continue; // if label contains class_id more than number of classes in the cfg-file
                        }
                        if(!truth.x) break; //不允许groundtruth的中心点为0!!
                        float iou = box_iou(pred, truth); //计算iou,注意(box.x,box.y)是中心点位置
                        if (iou > best_iou) {
                            best_iou = iou;
                            best_t = t; 
                        }
                    }
                    int obj_index = entry_index(l, b, n*l.w*l.h + j*l.w + i, 4); //读取预测的objectness索引
                    avg_anyobj += l.output[obj_index];//读取预测的objectness值,avg_anyobj训练状态检测量
                    /*
                    objectness 学习
                    iou < l.ignore_thresh: 作为负样本
                    iou > l.truth_thresh: 作为正样本
                    其他:不参与训练
                    */
                    l.delta[obj_index] = 0 - l.output[obj_index]; //默认作为负样本,目标objectness=0,误差设置为0 - l.output[obj_index]
                    if (best_iou > l.ignore_thresh) {
                        l.delta[obj_index] = 0; //iou较大,不能作为负样本, 清除误差
                    }
                    if (best_iou > l.truth_thresh) {
                        l.delta[obj_index] = 1 - l.output[obj_index]; //iou足够大,是正样本,目标是objectness=1, 梯度设置为1 - l.output[obj_index]

                        int class_id = state.truth[best_t*(4 + 1) + b*l.truths + 4];//groundtruth classid
                        if (l.map) class_id = l.map[class_id];
                        int class_index = entry_index(l, b, n*l.w*l.h + j*l.w + i, 4 + 1); //预测的classid
                        //计算classid的误差(one-hot-encoding,计算方法和objectness类似,但这里还支持focal_loss)
                        delta_yolo_class(l.output, l.delta, class_index, class_id, l.classes, l.w*l.h, 0, l.focal_loss);
                        box truth = float_to_box_stride(state.truth + best_t*(4 + 1) + b*l.truths, 1);
                        //计算box误差 (倒数第二个参数的意思是?)
                        delta_yolo_box(truth, l.output, l.biases, l.mask[n], box_index, i, j, l.w, l.h, state.net.w, state.net.h, l.delta, (2-truth.w*truth.h), l.w*l.h);
                    }
                }
            }
        } //end of "for (j = 0; j < l.h; ++j)"

        //下面这段代码遍历每个groundtruth object, 查找iou最大的anchor
        //这和上面遍历所有grid的anchor,查找iou最大的groundtruth object作用一样,但如此这样可以避免anchor设置不合适,没有完全覆盖
        //groundtruth object的情况?
        for(t = 0; t < l.max_boxes; ++t){
            box truth = float_to_box_stride(state.truth + t*(4 + 1) + b*l.truths, 1);
            int class_id = state.truth[t*(4 + 1) + b*l.truths + 4];
            if (class_id >= l.classes) continue; // if label contains class_id more than number of classes in the cfg-file

            if(!truth.x) break;
            float best_iou = 0;
            int best_n = 0;
            i = (truth.x * l.w);
            j = (truth.y * l.h);
            box truth_shift = truth;
            truth_shift.x = truth_shift.y = 0;
            for(n = 0; n < l.total; ++n){ //所有anchor都参与计算iou
                box pred = {0}; //中心位置都变成左上角
                pred.w = l.biases[2*n]/ state.net.w; //按网络输入尺寸归一化
                pred.h = l.biases[2*n+1]/ state.net.h;
                float iou = box_iou(pred, truth_shift);
                if (iou > best_iou){
                    best_iou = iou;
                    best_n = n;
                }
            }

            int mask_n = int_index(l.mask, best_n, l.n); //只有在mask中指定的anchor进行如下计算
            if(mask_n >= 0){ 
                int box_index = entry_index(l, b, mask_n*l.w*l.h + j*l.w + i, 0);
                //和前面的计算一样,但读取了返回值iou
                float iou = delta_yolo_box(truth, l.output, l.biases, best_n, box_index, i, j, l.w, l.h, state.net.w, state.net.h, l.delta, (2-truth.w*truth.h), l.w*l.h);

                int obj_index = entry_index(l, b, mask_n*l.w*l.h + j*l.w + i, 4);
                avg_obj += l.output[obj_index];
                l.delta[obj_index] = 1 - l.output[obj_index];

                int class_id = state.truth[t*(4 + 1) + b*l.truths + 4];
                if (l.map) class_id = l.map[class_id];
                int class_index = entry_index(l, b, mask_n*l.w*l.h + j*l.w + i, 4 + 1);
                //和前面的计算一样,但读取了一个状态信息avg_cat
                delta_yolo_class(l.output, l.delta, class_index, class_id, l.classes, l.w*l.h, &avg_cat, l.focal_loss);

                ++count;
                ++class_count;
                if(iou > .5) recall += 1;
                if(iou > .75) recall75 += 1;
                avg_iou += iou;
            }
        }
    }
    //avg_iou和avg_cat如果差异过大,是不是说明配置有问题?
    *(l.cost) = pow(mag_array(l.delta, l.outputs * l.batch), 2);
    printf("Region %d Avg IOU: %f, Class: %f, Obj: %f, No Obj: %f, .5R: %f, .75R: %f,  count: %d\n", state.index, avg_iou/count, avg_cat/class_count, avg_obj/count, avg_anyobj/(l.w*l.h*l.n*l.batch), recall/count, recall75/count, count);
}